On Tue, Dec 6, 2016 at 3:35 PM, Ben Pfaff <[email protected]> wrote: > On Tue, Dec 06, 2016 at 02:45:19PM -0500, Russell Bryant wrote: > > On Mon, Dec 5, 2016 at 2:22 AM, Ben Pfaff <[email protected]> wrote: > > > > > On Sun, Dec 04, 2016 at 04:13:44PM -0800, Darrell Ball wrote: > > > > This patch adds datapaths of interest support where only datapaths of > > > > local interest are monitored by the ovn-controller ovsdb client. The > > > > idea is to do a flood fill in ovn-controller of datapath associations > > > > calculated by northd. A new column is added to the SB database > > > > datapath_binding table - related_datapaths to facilitate this so all > > > > datapaths associations are known quickly in ovn-controller. This > > > > allows monitoring to adapt quickly with a single new monitor setting > > > > for all datapaths of interest locally. > > > > > > Hi Darrell, the series I just sent out has some relevance here. It > > > makes ovn-controller only implement the datapaths and ports flows that > > > are relevant to a given hypervisor, even though it does not affect what > > > part of the database is replicated. The particularly relevant patch is > > > this: > > > https://patchwork.ozlabs.org/patch/702608/ > > > > > > I suggest having a look at the series. > > > > > > > I'm in the middle of doing some control plane performance testing using > > OpenStack and OVN. I may have time to do another run with some > additional > > patches applied to see how they affect performance. Ideally I'd be able > to > > do that this week, though. > > > > If I'm able to fit this in, what do you guys suggest if I've only got one > > shot? This patch or Ben's series? > > This series is aimed more at technical debt and code clarity than at > performance. It might help with performance, but that's not the main > goal. > > Darrell/Liran's series is motivated by performance. If it doesn't help > performance, then probably it shouldn't be applied because it makes the > code harder to understand. (Disclaimer: I haven't read the recent > versions.) >
I just deployed this change in my environment and was not able to observe a noticeable difference in CPU consumption of ovn-controller. (ovn-northd was updated, as well.) The scenario was creating 500 VMs through OpenStack at a concurrency of 64 (up to 64 VMs at a time). Each VM also received its own network. The 500 VMs were distributed among 9 hypervisors. Once all 500 VMs were up, they were all deleted. The CPU consumption looked roughly the same both before and after the patch and time to complete the test scenario was also similar. ovn-controller hits 100% CPU about 30% of the way into creating VMs and stays there until we finish creating them and start deleting them instead. -- Russell Bryant _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
